A Fast Parallel Treecode with GRAPE
Junichiro Makino · Publications of the Astronomical Society of Japan · 2004
Abstract In this paper, we describe an implementation of the Barnes–Hut treecode on a cluster of PCs (Beowulf-class machines) equipped with GRAPE hardware. Our implementation is designed to achieve good performance on clusters of GRAPE with a relatively slow network, such as the standard Gigabit Ethernet. With 12 processors, the speedup over single-processor code reaches around 10. Our code is available on the web.